Computer aided design system for predicting energetic molecule based on machine learning performance

Abstract

The invention discloses a computer aided design system for predicting energetic molecules based on machine learning performance, and belongs to the technical field of computer aided design systems. The system comprises: a molecule rapid generation module, which is used for generating all possible molecular structural formulas according to permutation and combination of a molecular mother ring anda substituent group input by a user, and removing a repeated structure; a molecular data set module used for recording reported energetic molecular performance data, and the energetic molecular performance data being mainly used as a training set of the machine learning model training module; a molecular descriptor generation module used for calculating molecular descriptors of molecules generatedby the molecular rapid generation module or molecules in the molecular data set module or molecules input by a user; a machine learning model training module used for training and storing a model byadopting a machine learning algorithm according to the data of the molecular descriptor set; and a performance prediction module used for reading the model and carrying out performance prediction on the molecules generated by the molecule rapid generation module or the molecules input by the user.

Description

technical field

[0001] The invention relates to a computer-aided design system, in particular to a computer-aided design system for predicting the properties of energetic molecules based on machine learning methods. The system can assist researchers to generate all possible molecular structures according to specific parent rings and substituent arrangements, and realize performance prediction based on machine learning methods. Background technique

[0002] If Nobel's development of "Dynamat" is taken as a starting point, the development of modern energetic materials has a history of nearly 200 years. During this period, researchers successively discovered many new types of high-performance energetic molecules, such as RDX, HMX, CL-20, etc., but more attempts ended in failure, and the design and development of energetic molecules still need to consume a lot of energy. time and labor costs. The reason is that, on the one hand, the mainstream research and development model of...

Examples

Embodiment 1

[0050] Figure 5 is a process for predicting the performance data of a single known energetic molecule RDX. The first step is to input the molecular structure of RDX represented by the SMILES code, the second step is to optimize the structure and calculate the descriptor, and the third step is to select the model generated by the Kernel Ridge Regression algorithm (KRR) to evaluate the density, detonation velocity and detonation pressure performance of RDX. predict. Compared with the experimental value, the error of the predicted result is very small.

Embodiment 2

[0052] Image 6 It is a flow for the whole molecular space generation and performance prediction based on benzene ring and amino and nitro substituents. The user only needs to input the structural formulas of benzene ring, nitro group and amino group expressed in SMILES format, and the rapid molecular generation module will automatically generate 91 molecular structures without repetition, and then the molecular descriptor module will optimize the 91 molecular structures and calculate the descriptors. Then, the system will predict the performance of these 91 molecules according to the performance that the user wants to predict and the model generated by the algorithm used, such as the Kernel Ridge Regression (KRR) model. The whole process only takes a few minutes, and the speed Far beyond traditional calculation methods, and still maintain high accuracy.

technical field

[0001] The invention relates to a computer-aided design system, in particular to a computer-aided design system for predicting the properties of energetic molecules based on machine learning methods. The system can assist researchers to generate all possible molecular structures according to specific parent rings and substituent arrangements, and realize performance prediction based on machine learning methods. Background technique

[0002] If Nobel's development of "Dynamat" is taken as a starting point, the development of modern energetic materials has a history of nearly 200 years. During this period, researchers successively discovered many new types of high-performance energetic molecules, such as RDX, HMX, CL-20, etc., but more attempts ended in failure, and the design and development of energetic molecules still need to consume a lot of energy. time and labor costs. The reason is that, on the one hand, the mainstream research and development model of...

Examples

Embodiment 1

[0050] Figure 5 is a process for predicting the performance data of a single known energetic molecule RDX. The first step is to input the molecular structure of RDX represented by the SMILES code, the second step is to optimize the structure and calculate the descriptor, and the third step is to select the model generated by the Kernel Ridge Regression algorithm (KRR) to evaluate the density, detonation velocity and detonation pressure performance of RDX. predict. Compared with the experimental value, the error of the predicted result is very small.

Embodiment 2

[0052] Image 6 It is a flow for the whole molecular space generation and performance prediction based on benzene ring and amino and nitro substituents. The user only needs to input the structural formulas of benzene ring, nitro group and amino group expressed in SMILES format, and the rapid molecular generation module will automatically generate 91 molecular structures without repetition, and then the molecular descriptor module will optimize the 91 molecular structures and calculate the descriptors. Then, the system will predict the performance of these 91 molecules according to the performance that the user wants to predict and the model generated by the algorithm used, such as the Kernel Ridge Regression (KRR) model. The whole process only takes a few minutes, and the speed Far beyond traditional calculation methods, and still maintain high accuracy.

Images